Temporal and spatial analysis of landslide-triggering rainfall conditions in Qinghai-Tibet Plateau, China

Abstract

Landslide on the Qinghai-Tibet Plateau (QTP) is expected to be more affected by climate change due to the sensitivity of this unique climatic and geomorphological area to variations in temperature and precipitation. As an important response signal to climate change, a systematic framework for the assessment of landslide hazard and risk in QTP is necessary to investigate the potential impacts of climate change on landslides and related exposures. The study aims to establish an integrated model that synthesizes spatial and temporal landslide prediction, using statistical analysis, machine learning, and quantitative methods. The temporal landslide prediction is made by means of empirical rainfall thresholds, based on satellite rainfall estimates, whose feasibility for defining landslide-triggering rainfall thresholds was proved by several studies. A well-documented hazard database of the QTP provided by the China Geological Survey (4519 records from 2001 to 2022) indicates that landslides occurred here are mostly induced by rainfall from April to October, with an obvious seasonal characteristic, resulting in fatalities, damage, and affected population. According to the database, 3542 landslides are associated to a rainfall trigger. Based on the satellite-based rainfall product of CHIRPS (Climate Hazards Group InfraRed Precipitation with Station data, version 2.0 final) daily data, we find that the rainfall of the occurrence day and the antecedent rainfall over the seven days before the landslides are significant indicators for the rainfall induced hazard. Using the frequentist method, the event duration-cumulated event rainfall (ED) thresholds at different non-exceedance probabilities for landslide triggering are calculated for the whole QTP area and for different environmental subdivisions within it. The thresholds show a robust definition with low parameter uncertainty. This is the first attempt to define empirical rainfall thresholds for landslide occurrence specifically for the QTP. Given the long-term of the used database, temporal and spatial analyses are conducted, to search for variations in the rainfall triggering conditions according to landslide locations and time of occurrence. Variations in the seasonal distribution and in the annual trends (using 5-year moving windows from 2007 to 2002) are evaluated. The impact of variations in rainfall patterns due to climate change making the landscape of the QTP more prone to landslides during the recent-most ten years is demonstrated by the gradual change of thresholds with lower intercepts and slopes. That means, for a certain rainfall duration, there is a tendency of lower rainfall threshold to trigger a landslide. The thresholds here defined are further combined with landslide susceptibility map based on Random Forest to derive a landslide hazard map for the interested area.